diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 229 |
1 files changed, 175 insertions, 54 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index b372d8d650a5..e28b73609fbc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -46,6 +46,7 @@ #include "amdgpu.h" #include "amdgpu_object.h" #include "amdgpu_trace.h" +#include "amdgpu_amdkfd.h" #include "bif/bif_4_1_d.h" #define DRM_FILE_PAGE_OFFSET (0x100000000ULL >> PAGE_SHIFT) @@ -203,6 +204,12 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, .flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM }; + if (bo->type == ttm_bo_type_sg) { + placement->num_placement = 0; + placement->num_busy_placement = 0; + return; + } + if (!amdgpu_ttm_bo_is_amdgpu_bo(bo)) { placement->placement = &placements; placement->busy_placement = &placements; @@ -213,9 +220,7 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, abo = ttm_to_amdgpu_bo(bo); switch (bo->mem.mem_type) { case TTM_PL_VRAM: - if (adev->mman.buffer_funcs && - adev->mman.buffer_funcs_ring && - adev->mman.buffer_funcs_ring->ready == false) { + if (!adev->mman.buffer_funcs_enabled) { amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU); } else if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size && !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) { @@ -260,6 +265,13 @@ static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp) { struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo); + /* + * Don't verify access for KFD BOs. They don't have a GEM + * object associated with them. + */ + if (abo->kfd_bo) + return 0; + if (amdgpu_ttm_tt_get_usermm(bo->ttm)) return -EPERM; return drm_vma_node_verify_access(&abo->gem_base.vma_node, @@ -331,7 +343,7 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, const uint64_t GTT_MAX_BYTES = (AMDGPU_GTT_MAX_TRANSFER_SIZE * AMDGPU_GPU_PAGE_SIZE); - if (!ring->ready) { + if (!adev->mman.buffer_funcs_enabled) { DRM_ERROR("Trying to move memory with ring turned off.\n"); return -EINVAL; } @@ -577,12 +589,9 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict, amdgpu_move_null(bo, new_mem); return 0; } - if (adev->mman.buffer_funcs == NULL || - adev->mman.buffer_funcs_ring == NULL || - !adev->mman.buffer_funcs_ring->ready) { - /* use memcpy */ + + if (!adev->mman.buffer_funcs_enabled) goto memcpy; - } if (old_mem->mem_type == TTM_PL_VRAM && new_mem->mem_type == TTM_PL_SYSTEM) { @@ -621,6 +630,7 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ { struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); + struct drm_mm_node *mm_node = mem->mm_node; mem->bus.addr = NULL; mem->bus.offset = 0; @@ -640,6 +650,15 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ /* check if it's visible */ if ((mem->bus.offset + mem->bus.size) > adev->gmc.visible_vram_size) return -EINVAL; + /* Only physically contiguous buffers apply. In a contiguous + * buffer, size of the first mm_node would match the number of + * pages in ttm_mem_reg. + */ + if (adev->mman.aper_base_kaddr && + (mm_node->size == mem->num_pages)) + mem->bus.addr = (u8 *)adev->mman.aper_base_kaddr + + mem->bus.offset; + mem->bus.base = adev->gmc.aper_base; mem->bus.is_iomem = true; break; @@ -674,7 +693,6 @@ struct amdgpu_ttm_gup_task_list { struct amdgpu_ttm_tt { struct ttm_dma_tt ttm; - struct amdgpu_device *adev; u64 offset; uint64_t userptr; struct mm_struct *usermm; @@ -832,6 +850,7 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm) static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem) { + struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); struct amdgpu_ttm_tt *gtt = (void*)ttm; uint64_t flags; int r = 0; @@ -858,9 +877,9 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm, return 0; } - flags = amdgpu_ttm_tt_pte_flags(gtt->adev, ttm, bo_mem); + flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem); gtt->offset = (u64)bo_mem->start << PAGE_SHIFT; - r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages, + r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages, ttm->pages, gtt->ttm.dma_address, flags); if (r) @@ -937,6 +956,7 @@ int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo) static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) { + struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); struct amdgpu_ttm_tt *gtt = (void *)ttm; int r; @@ -947,7 +967,7 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) return 0; /* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */ - r = amdgpu_gart_unbind(gtt->adev, gtt->offset, ttm->num_pages); + r = amdgpu_gart_unbind(adev, gtt->offset, ttm->num_pages); if (r) DRM_ERROR("failed to unbind %lu pages at 0x%08llX\n", gtt->ttm.ttm.num_pages, gtt->offset); @@ -968,22 +988,20 @@ static struct ttm_backend_func amdgpu_backend_func = { .destroy = &amdgpu_ttm_backend_destroy, }; -static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_bo_device *bdev, - unsigned long size, uint32_t page_flags, - struct page *dummy_read_page) +static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo, + uint32_t page_flags) { struct amdgpu_device *adev; struct amdgpu_ttm_tt *gtt; - adev = amdgpu_ttm_adev(bdev); + adev = amdgpu_ttm_adev(bo->bdev); gtt = kzalloc(sizeof(struct amdgpu_ttm_tt), GFP_KERNEL); if (gtt == NULL) { return NULL; } gtt->ttm.ttm.func = &amdgpu_backend_func; - gtt->adev = adev; - if (ttm_dma_tt_init(>t->ttm, bdev, size, page_flags, dummy_read_page)) { + if (ttm_sg_tt_init(>t->ttm, bo, page_flags)) { kfree(gtt); return NULL; } @@ -1009,7 +1027,8 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm, if (slave && ttm->sg) { drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages, - gtt->ttm.dma_address, ttm->num_pages); + gtt->ttm.dma_address, + ttm->num_pages); ttm->state = tt_unbound; return 0; } @@ -1167,6 +1186,23 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, { unsigned long num_pages = bo->mem.num_pages; struct drm_mm_node *node = bo->mem.mm_node; + struct reservation_object_list *flist; + struct dma_fence *f; + int i; + + /* If bo is a KFD BO, check if the bo belongs to the current process. + * If true, then return false as any KFD process needs all its BOs to + * be resident to run successfully + */ + flist = reservation_object_get_list(bo->resv); + if (flist) { + for (i = 0; i < flist->shared_count; ++i) { + f = rcu_dereference_protected(flist->shared[i], + reservation_object_held(bo->resv)); + if (amdkfd_fence_check_mm(f, current->mm)) + return false; + } + } switch (bo->mem.mem_type) { case TTM_PL_TT: @@ -1306,11 +1342,12 @@ static int amdgpu_ttm_fw_reserve_vram_init(struct amdgpu_device *adev) if (adev->fw_vram_usage.size > 0 && adev->fw_vram_usage.size <= vram_size) { - r = amdgpu_bo_create(adev, adev->fw_vram_usage.size, - PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, - AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | - AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, NULL, NULL, - &adev->fw_vram_usage.reserved_bo); + r = amdgpu_bo_create(adev, adev->fw_vram_usage.size, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | + AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, + ttm_bo_type_kernel, NULL, + &adev->fw_vram_usage.reserved_bo); if (r) goto error_create; @@ -1402,7 +1439,11 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) adev->gmc.visible_vram_size = vis_vram_limit; /* Change the size here instead of the init above so only lpfn is affected */ - amdgpu_ttm_set_active_vram_size(adev, adev->gmc.visible_vram_size); + amdgpu_ttm_set_buffer_funcs_status(adev, false); +#ifdef CONFIG_64BIT + adev->mman.aper_base_kaddr = ioremap_wc(adev->gmc.aper_base, + adev->gmc.visible_vram_size); +#endif /* *The reserved vram for firmware must be pinned to the specified @@ -1495,6 +1536,9 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev) amdgpu_ttm_debugfs_fini(adev); amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL); amdgpu_ttm_fw_reserve_vram_fini(adev); + if (adev->mman.aper_base_kaddr) + iounmap(adev->mman.aper_base_kaddr); + adev->mman.aper_base_kaddr = NULL; ttm_bo_clean_mm(&adev->mman.bdev, TTM_PL_VRAM); ttm_bo_clean_mm(&adev->mman.bdev, TTM_PL_TT); @@ -1510,18 +1554,30 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev) DRM_INFO("amdgpu: ttm finalized\n"); } -/* this should only be called at bootup or when userspace - * isn't running */ -void amdgpu_ttm_set_active_vram_size(struct amdgpu_device *adev, u64 size) +/** + * amdgpu_ttm_set_buffer_funcs_status - enable/disable use of buffer functions + * + * @adev: amdgpu_device pointer + * @enable: true when we can use buffer functions. + * + * Enable/disable use of buffer functions during suspend/resume. This should + * only be called at bootup or when userspace isn't running. + */ +void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) { - struct ttm_mem_type_manager *man; + struct ttm_mem_type_manager *man = &adev->mman.bdev.man[TTM_PL_VRAM]; + uint64_t size; - if (!adev->mman.initialized) + if (!adev->mman.initialized || adev->in_gpu_reset) return; - man = &adev->mman.bdev.man[TTM_PL_VRAM]; /* this just adjusts TTM size idea, which sets lpfn to the correct value */ + if (enable) + size = adev->gmc.real_vram_size; + else + size = adev->gmc.visible_vram_size; man->size = size >> PAGE_SHIFT; + adev->mman.buffer_funcs_enabled = enable; } int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma) @@ -1620,6 +1676,11 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, unsigned i; int r; + if (direct_submit && !ring->ready) { + DRM_ERROR("Trying to move memory with ring turned off.\n"); + return -EINVAL; + } + max_bytes = adev->mman.buffer_funcs->copy_max_bytes; num_loops = DIV_ROUND_UP(byte_count, max_bytes); num_dw = num_loops * adev->mman.buffer_funcs->copy_num_dw; @@ -1693,7 +1754,7 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo, struct amdgpu_job *job; int r; - if (!ring->ready) { + if (!adev->mman.buffer_funcs_enabled) { DRM_ERROR("Trying to clear memory with ring turned off.\n"); return -EINVAL; } @@ -1929,38 +1990,98 @@ static const struct file_operations amdgpu_ttm_gtt_fops = { #endif -static ssize_t amdgpu_iova_to_phys_read(struct file *f, char __user *buf, - size_t size, loff_t *pos) +static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf, + size_t size, loff_t *pos) { struct amdgpu_device *adev = file_inode(f)->i_private; - int r; - uint64_t phys; struct iommu_domain *dom; + ssize_t result = 0; + int r; - // always return 8 bytes - if (size != 8) - return -EINVAL; + dom = iommu_get_domain_for_dev(adev->dev); - // only accept page addresses - if (*pos & 0xFFF) - return -EINVAL; + while (size) { + phys_addr_t addr = *pos & PAGE_MASK; + loff_t off = *pos & ~PAGE_MASK; + size_t bytes = PAGE_SIZE - off; + unsigned long pfn; + struct page *p; + void *ptr; + + bytes = bytes < size ? bytes : size; + + addr = dom ? iommu_iova_to_phys(dom, addr) : addr; + + pfn = addr >> PAGE_SHIFT; + if (!pfn_valid(pfn)) + return -EPERM; + + p = pfn_to_page(pfn); + if (p->mapping != adev->mman.bdev.dev_mapping) + return -EPERM; + + ptr = kmap(p); + r = copy_to_user(buf, ptr, bytes); + kunmap(p); + if (r) + return -EFAULT; + + size -= bytes; + *pos += bytes; + result += bytes; + } + + return result; +} + +static ssize_t amdgpu_iomem_write(struct file *f, const char __user *buf, + size_t size, loff_t *pos) +{ + struct amdgpu_device *adev = file_inode(f)->i_private; + struct iommu_domain *dom; + ssize_t result = 0; + int r; dom = iommu_get_domain_for_dev(adev->dev); - if (dom) - phys = iommu_iova_to_phys(dom, *pos); - else - phys = *pos; - r = copy_to_user(buf, &phys, 8); - if (r) - return -EFAULT; + while (size) { + phys_addr_t addr = *pos & PAGE_MASK; + loff_t off = *pos & ~PAGE_MASK; + size_t bytes = PAGE_SIZE - off; + unsigned long pfn; + struct page *p; + void *ptr; + + bytes = bytes < size ? bytes : size; - return 8; + addr = dom ? iommu_iova_to_phys(dom, addr) : addr; + + pfn = addr >> PAGE_SHIFT; + if (!pfn_valid(pfn)) + return -EPERM; + + p = pfn_to_page(pfn); + if (p->mapping != adev->mman.bdev.dev_mapping) + return -EPERM; + + ptr = kmap(p); + r = copy_from_user(ptr, buf, bytes); + kunmap(p); + if (r) + return -EFAULT; + + size -= bytes; + *pos += bytes; + result += bytes; + } + + return result; } -static const struct file_operations amdgpu_ttm_iova_fops = { +static const struct file_operations amdgpu_ttm_iomem_fops = { .owner = THIS_MODULE, - .read = amdgpu_iova_to_phys_read, + .read = amdgpu_iomem_read, + .write = amdgpu_iomem_write, .llseek = default_llseek }; @@ -1973,7 +2094,7 @@ static const struct { #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS { "amdgpu_gtt", &amdgpu_ttm_gtt_fops, TTM_PL_TT }, #endif - { "amdgpu_iova", &amdgpu_ttm_iova_fops, TTM_PL_SYSTEM }, + { "amdgpu_iomem", &amdgpu_ttm_iomem_fops, TTM_PL_SYSTEM }, }; #endif |